In the pre-training stage, some parameters of the model are frozen, focusing on training the segmentation decoder, classification layer, and positioning structure to identify tampered areas and types ...